Who is responsible for the output of generative LLMs? Is it the person running the model, the person or group or company which trained the model, the creator of the initial parameters, the person who regurgitates what the model produced?
Whether the Xerox photocopier, the Sony Betamax or (so far) generative AI, judges are ruling that the fault in using technology to infringe copyright doesn't lie in the technology, but the way the technology is used by a person.
Generative AI isn't (yet) a person, so it is neither capable of creating copyrightable works, nor in itself infringing copyright.
I wrote a research paper on it, if you're interested.